# considerations in highly availability design

+ Network latency
  + The Raft protocol requires that writes be replicated to at least 2 nodes. (three replica)
  + The Leader may not be in the same region as the TiDB Server that initiates the read
  + Read to access PD to get TSO once, transaction to get 2 times
+ The Raft protocol itself
  + The Raft protocol requires that writes be replicated to at least 2 nodes. (three replica)
  + Preferably an odd number of copies
  + The distribution of replica is best combined with the distribution of TiKV nodes
+ other
  + Requirement for extra activity
------------------------------------------------------------
[no notes]